Based on promising preclinical results, a variety of TGF-beta pathway antagonists are in early phase clinical trials for the treatment of advanced cancer. However, given the complex biology of TGF-beta, the successful development of TGF-beta antagonists for cancer therapy will depend on a clear understanding of how these agents work, and the related question of how to select patients who will benefit from this type of treatment. Using a panel of 12 mouse syngeneic allograft models of metastatic breast cancer, with metastatic burden as the primary endpoint, we uncovered heterogeneous responses to TGF-beta antagonism. TGF-beta pathway blockade inhibited metastasis in some models, while having no effect on or stimulating metastasis in other models. We are continuing to apply targeted and discovery-based approaches to address molecular and biological mechanisms underlying the heterogeneity of therapeutic response and to generate useful predictive biomarkers. To do this, we have extensively characterized the panel of metastatic models with respect to genomic, transcriptomic and clinical features, and we have made this information publically available. Using this information we have shown that plausible candidate biomarkers suggested by the existing literature (eg. p53 mutation status, claudin-low tumor phenotype) do not appear to correlate with response to TGF-beta antagonism. Unexpectedly, neither TGF-beta expression levels nor extent of TGF-beta pathway activation in the primary tumor predict response to anti-TGF-beta therapy. However, tumors from models showing a desirable response to TGF-beta antagonism are characterized by transcriptomic evidence of TGF-beta pathway activation in the untreated state. Thus in vivo, the output of TGF-beta signaling may be a more sensitive indicator of the activation status of the TGF-beta pathway than any of the input parameters such as ligand levels and status of signal transduction components. We have continued to explore the mechanisms underlying the undesirable prometastatic responses to TGF-beta antagonists and we have shown that they are independent of the immune system and appear to involve disruption of direct metastasis-suppressing effects of TGF-beta on the tumor cells themselves. Analysis of in vitro responses to TGF-beta across the model panel led us to hypothesize that the metastasis-suppressing effect of TGF-beta is targeted specifically to the cancer stem cell subpopulation, and we have preliminary evidence that this is also true in vivo. We are also continuing to address ways of improving the therapeutic efficacy of TGF-beta antagonism. In one approach, we are using isoform-selective neutralizing antibodies to determine whether all three TGF-beta isoforms should be targeted for optimal response. We have generated substantial correlative data to suggest that TGF-beta1 and TGF-beta3 may play opposing roles in breast cancer. Using TGF-beta isoform-selective antibodies in preclinical mouse models, we find that sparing TGF-beta3 results in an altered metabolic profile and upregulation of genes that are associated with good prognosis in human breast cancer datasets, suggesting that this approach may have therapeutic benefit.